Prompt Injection
Prompt injection is an attack where a user (or data) inserts instructions into a prompt so the model follows those instructions instead of or in addition to the intended task. It can leak system prompts, change behavior, or cause misuse.
In Simple Terms
Think of it as slipping a fake script to an actor: the model may follow the injected lines instead of the intended script.
Detailed Explanation
Injection can be direct (user types “ignore previous instructions”) or indirect (malicious content in a retrieved document or web page that the model is asked to summarize). Defenses include input sanitization, least-privilege system prompts, output checks, and guardrails. As more apps embed LLMs with access to tools and data, prompt injection is a growing security concern. Designers should assume that some inputs will be adversarial and limit what the model can do and see.
Related Terms
Artificial Intelligence
The simulation of human intelligence processes by machines, especially computer systems.
Read moreMachine Learning
A subset of AI that enables systems to learn and improve from experience without being explicitly programmed.
Read moreBias in AI
Bias in AI is systematic error or unfairness in how a model treats individuals or groups, often reflecting skewed data or flawed design. It can worsen existing inequalities if left unchecked.
Read moreWant to Implement AI in Your Business?
Let's discuss how these AI concepts can drive value in your organization.
Schedule a Consultation